A test of mental arithmetic skills was developed and normed for children in grades one to six. Statistical techniques were used to predict the probability correct for each item and to rank order the items. Students in a given grade took only one level of the test and inferences were made of their performance on those tests which they did not take. Using performance models, student scores on each test were predicted by estimating parameters in functional relationships between probability correct for the population and the individual. A function was found mapping the probability correct for a given item on one test and grade to the probability correct for the same item on another test and grade. Then the probability correct on any item in any test could be predicted for a given grade. Performance models for each student were used to predict student scores on each test. A functional relationship was found between the probability correct for the population and the probability correct for the individual on groups of items. From the student's performance curves, the probability correct for those items that he or she did not take could be predicted, and a total score on any given test thus computed. Traditional norming procedures were then applied to these scores in determining grade equivalents. (Author/RC)